API Reference

Complete reference for all HTTP and SSE endpoints provided by the LLM Telemetry Proxy (:9090) and the Dashboard Server (:9118).

---

⚡ Proxy Endpoints (Port 9090)

Inference Gateway

Method	Path	Description
POST	/v1/chat/completions	Proxies OpenAI chat completion requests with stream timing.
POST	/v1/completions	Proxies legacy text completions.
POST	/v1/embeddings	Proxies text embeddings.
GET	/v1/models	Lists upstream available models (passed through).
GET	/v1/models/{model}	Fetches model metadata.

---

Proxy Health & Token Budget

GET /health

Returns proxy health status, semaphore utilization, rolling 24-hour token budget, and upstream server status.

Example Response:

{
  "status": "ok",
  "upstream": "https://llm.ai.e-infra.cz/v1",
  "db": "/path/to/data/llm_telemetry.db",
  "rate_limiter": {
    "max_concurrent": 4,
    "active": 1,
    "queued": 0,
    "slot_cooldown_ms": 50,
    "total_admitted": 124,
    "total_queued": 15,
    "total_retries_429": 0,
    "peak_active": 4
  },
  "token_budget": {
    "daily_limit": 480000000,
    "total_used": 1450230,
    "remaining": 478549770,
    "percentage_used": 0.3
  }
}

---

Raw Payload Inspector Endpoints

Method	Path	Description
GET	/v1/raw-log/status	Returns whether raw payload logging is enabled and file size.
POST	/v1/raw-log/toggle	Toggles raw logging ON/OFF. Body: {"enabled": true}.
GET	/v1/raw-log/recent	Returns the last N raw payload JSON records. Query: ?limit=50.
POST	/v1/raw-log/clear	Truncates the logger/payloads.jsonl file.
GET	/v1/raw-log/stream	Server-Sent Events (SSE) live feed of raw incoming/outgoing payloads.

---

📊 Dashboard API Endpoints (Port 9118)

Telemetry Query API

GET /api/query (or /api/stats)

Executes parameterized aggregations against the telemetry SQLite database.

Query Parameters: - from (string, ISO-8601): Start timestamp filter. - to (string, ISO-8601): End timestamp filter. - model (string, repeatable): Filter by one or more model names. - call_type (string, repeatable): chat, embedding, rerank, props. - group_by (string): model, hour, day, call_type. - errors_only (boolean): true to filter only failed requests. - limit (integer): Maximum records when returning raw call lists (default 1000).

---

Cost & Model Configuration

GET /api/costs

Returns the active data/model_costs.json pricing tier configuration.

POST /api/costs/sync

Triggers an automated background pull from the LiteLLM pricing dataset, updating data/model_costs.json with dynamic tiers.

---

Process Lifecycle Control

Method	Path	Description
GET	/api/proxy/status	Returns running state, PID, port, and health check of the proxy gateway.
POST	/api/proxy/start	Spawns the proxy gateway as a background process.
POST	/api/proxy/stop	Terminates the running proxy process.
POST	/api/proxy/restart	Gracefully restarts the proxy process.
GET	/api/proxy/logs	Reads recent stdout/stderr output lines from data/proxy.log.
POST	/api/proxy/clear-logs	Clears data/proxy.log.
POST	/api/db/compress	Triggers the database compression script proxy/db_compress.py.